Large Deation analysis for ATM Traffic

ATM(Asynchrpnous Transfer Model，异步传输模式)网络承载着数字化语音、电影、编码视频和数据等多种类型的业务。近年来，许多研究人员报道，ATM业务在这样的网络系统中具有非常复杂的特性-混沌、自相似、长期依赖和多样化的服务质量(Qos)要求。通常，需要将缓冲器溢出引起的分组丢失概率控制在非常小的水平以下，例如，在10-9量级。确定罕见事件的概率变得非常重要。作为一种估计罕见事件概率的有效方法，大偏差理论和方法受到了广泛的关注。该研究考虑了一个由两个并行队列组成的ATM网络，它对于具有两种不同类型的业务的通信系统的建模非常重要：实时业务(如语音和视频)和非实时业务(如数据)。在…环境下，我们将系统作为轮询模型进行分析(I)到达过程为泊松过程，服务时间服从指数分布。(1)在(I)非抢占优先级和(Ii)抢占优先级下，m个服务器根据(M，N)门限服务调度为两个队列服务。(2)两个服务器根据(I)基于门限的控制服务调度和(Ii)延迟控制服务调度为两个队列服务。(3)在一般服务分配下，一个服务台按照(M，N)门限服务调度为两个队列服务。我们推导了平稳联合队长分布的母函数，并得到了分组丢失概率。(Ii)利用大偏差技术，我们得到了具有马尔可夫调制到达过程和Bernoulli服务调度的离散时间流体轮询系统的分组丢失概率的上界和下界。(Iii)利用(Ii)中得到的分组丢失概率的上下界，我们给出了一个具有一般随机过程和多服务台的轮询系统的接纳控制算法。较少

ATM(Asynchrpnous Transfer Mode) networks accommodate various types of traffic such as digitized voice, movie, encoded video and data, etc. Recently, it has been reported by many researchers that ATM traffics in such network systems has very complicated properties -chaos, self-similarity, long-terms dependence and diverse quality of service (QoS) requirements. In general, the packet loss probabilities due to the buffer overflow are required to control below very small level, e.g., in the order of 10-9. Determining the rare event probability becomes very important. As a powerful technique in estimating rare event probability, the large deviation theory and approach have received remarkable attention.This study consider an ATM network consisting of two-parallel queue which is important in modeling a communication system with two different types of the traffic: real-time traffic(such as voice and video) and non-real-time traf」Lc(such as data). We analyzed the systems as polling models unde … More r the.various service schedules.(I) The arrival processes are Poisson processes, and the service time distributions are exponential. (1) m servers serve the two queues according to an (M,N)-threshold service schedule under the (I) non-preemptive priority and (ii) preemptive priority. (2) Two servers serve the two queues according to (I) Threshould-based control service schedule and (ii) Hyteretic control service schedule. (3) Under the general service distribution, one server serves the two queues according to an (M,N)-threshold service schedule. We derived the generating functions of the stationary joint queue-length distributions, and obtain the packet loss probabilities.(II) Using large deviation technique, we derive the upper and lower bounds of the packet loss probabilities for a discrete-time fluid polling system with Markov-modulated arrival processes and Bernoulli service schedule.(III) Using the upper and lower bounds of the packet loss probabilities obtained in (II), we present an admission control algorithm for a polling system with general arival processes and multi-server. Less